Image credit: Christine Ahrends (CC BY 4.0)
Watching how people behave over time can provide insights into their personality, mental health, as well as how they think and problem solve. Like behaviour, brain activity patterns constantly change, both at rest and in response to external events. These changes might reveal crucial information about a person that cannot be seen when looking at a single snapshot or an average of brain activity.
It has been difficult for researchers to predict individual traits from the overarching dynamic patterns of brain activity measured using brain scans and other imaging tools. This is due to the patterns being too complex to be analyzed directly. Mathematical models like the Hidden Markov Model can describe dynamic patterns in brain activity, such as how different brain areas’ activity and interaction with one another changes over time. To use this type of model to predict individual traits, Ahrends et al. combined it with a machine learning technique known as the Fisher kernel.
Using this combination of techniques to model dynamic patterns of brain activity based on scans from 1,000 resting people allowed the researchers to successfully predict an individual’s age and their score in various cognitive tests. This approach was shown to more accurately predict traits than alternative methods.
In the future, researchers may use this new modeling technique to search for markers of disease in dynamic brain activity patterns. For example, this could provide information about the progression of neuropsychiatric diseases over time. It may also help neuroscientists study how dynamic brain activity patterns contribute to individual cognitive performance.
